2026.06.29 | 物理强化破解模拟瓶颈；手腕平移架起人机桥梁

【赞助商】OpenClaw快报每天五分钟，听听 OpenClaw 快报，带你了解最新动态和业内讨论传送门 https://www.xiaoyuzhoufm.com/podcast/6a1732a2dffa135d0ab5ef43【目录】本期的 15 篇论文如下：[00:31] 🤖 PhysisForcing: Physics Reinforced World Simulator for Robotic Manipulation（物理强化：面向机器人操作的物理增强世界模拟器）[01:28] 🤖 Translation as a Bridging Action: Transferring Manipulation Skills from Humans to Robots（翻译作为桥梁动作：将操作技能从人类迁移到机器人）[02:13] 🎨 Qwen-Image-2.0-RL Technical Report（Qwen-Image-2.0-RL技术报告）[03:01] 🔑 MultiHashFormer: Hash-based Generative Language Models（MultiHashFormer：基于哈希的自回归语言模型）[03:51] 🧠 Formalizing Latent Thoughts: Four Axioms of Thought Representation in LLMs（形式化潜在思维：大语言模型中思想表示的四条公理）[04:34] 🛡 SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning（SingGuard：一种具有动态推理能力的策略自适应多模态大语言模型护栏）[05:30] 🔍 ProMSA:Progressive Multimodal Search Agents for Knowledge-Based Visual Question Answering（ProMSA：渐进式多模态搜索代理用于基于知识的视觉问答）[06:36] 🛡 The Tatoxa System for Text Detoxification in Low-Resource Languages: The Case of Tatar（Tatoxa系统用于低资源语言文本去毒化：以鞑靼语为例）[07:30] 🤖 SimFoundry: Modular and Automated Scene Generation for Policy Learning and Evaluation（SimFoundry：用于策略学习与评估的模块化自动化场景生成）[08:30] 🧠 GBC: Gradient-Based Connections for Optimizing Multi-Agent Systems（GBC：基于梯度的连接优化多智能体系统）[09:31] 👕 Learning to Fold: prizewinning solution at LeHome Challenge 2026 (1st place online, 2nd offline)（学习折叠：LeHome挑战赛2026的获奖解决方案（线上第一名，线下第二名））[10:27] 🔍 Ko-WideSearch: A Korean Breadth-Search Benchmark for Exhaustive Set Enumeration by Web Agents（Ko-WideSearch：用于网络代理穷举集合枚举的韩语广度搜索基准）[11:20] 🗣 Thinking While Speaking: Inference-Time Knowledge Transfer for Responsive and Intelligent Conversational Voice Agents（边思考边说话：面向响应式与智能对话语音代理的推理时知识迁移）[12:14] 🎨 Parallel Rollout Approximation for Pixel-Space Autoregressive Image Generation（像素空间自回归图像生成的并行 rollout 近似方法）[13:06] 🛡 NormGuard: Reward-Preserving Norm Constraints in Flow-Matching Reinforcement Learning（NormGuard：流匹配强化学习中保持奖励的范数约束）【关注我们】您还可以在以下平台找到我们，获得播客内容以外更多信息小红书: AI速递在小宇宙查看该单集文稿